Deep Learning : Principles and Practices - CSE1016 - L33 + L34
Name: Nikhil V
Registration No: CH.EN.U4AIE22038
Lab - 7 : Transfer Learning and Fine Tuning on Plant Village Dataset
In [1]:
!pip install split-folders
Collecting split-folders Downloading split_folders-0.5.1-py3-none-any.whl.metadata (6.2 kB) Downloading split_folders-0.5.1-py3-none-any.whl (8.4 kB) Installing collected packages: split-folders Successfully installed split-folders-0.5.1
In [2]:
!pip install tensorflow
Requirement already satisfied: tensorflow in /opt/conda/lib/python3.10/site-packages (2.16.1) Requirement already satisfied: absl-py>=1.0.0 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (1.4.0) Requirement already satisfied: astunparse>=1.6.0 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (1.6.3) Requirement already satisfied: flatbuffers>=23.5.26 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (24.3.25) Requirement already satisfied: gast!=0.5.0,!=0.5.1,!=0.5.2,>=0.2.1 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (0.5.4) Requirement already satisfied: google-pasta>=0.1.1 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (0.2.0) Requirement already satisfied: h5py>=3.10.0 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (3.11.0) Requirement already satisfied: libclang>=13.0.0 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (18.1.1) Requirement already satisfied: ml-dtypes~=0.3.1 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (0.3.2) Requirement already satisfied: opt-einsum>=2.3.2 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (3.3.0) Requirement already satisfied: packaging in /opt/conda/lib/python3.10/site-packages (from tensorflow) (21.3) Requirement already satisfied: protobuf!=4.21.0,!=4.21.1,!=4.21.2,!=4.21.3,!=4.21.4,!=4.21.5,<5.0.0dev,>=3.20.3 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (3.20.3) Requirement already satisfied: requests<3,>=2.21.0 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (2.32.3) Requirement already satisfied: setuptools in /opt/conda/lib/python3.10/site-packages (from tensorflow) (70.0.0) Requirement already satisfied: six>=1.12.0 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (1.16.0) Requirement already satisfied: termcolor>=1.1.0 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (2.4.0) Requirement already satisfied: typing-extensions>=3.6.6 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (4.12.2) Requirement already satisfied: wrapt>=1.11.0 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (1.16.0) Requirement already satisfied: grpcio<2.0,>=1.24.3 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (1.64.1) Requirement already satisfied: tensorboard<2.17,>=2.16 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (2.16.2) Requirement already satisfied: keras>=3.0.0 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (3.3.3) Requirement already satisfied: tensorflow-io-gcs-filesystem>=0.23.1 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (0.37.0) Requirement already satisfied: numpy<2.0.0,>=1.23.5 in /opt/conda/lib/python3.10/site-packages (from tensorflow) (1.26.4) Requirement already satisfied: wheel<1.0,>=0.23.0 in /opt/conda/lib/python3.10/site-packages (from astunparse>=1.6.0->tensorflow) (0.43.0) Requirement already satisfied: rich in /opt/conda/lib/python3.10/site-packages (from keras>=3.0.0->tensorflow) (13.7.1) Requirement already satisfied: namex in /opt/conda/lib/python3.10/site-packages (from keras>=3.0.0->tensorflow) (0.0.8) Requirement already satisfied: optree in /opt/conda/lib/python3.10/site-packages (from keras>=3.0.0->tensorflow) (0.11.0) Requirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.10/site-packages (from requests<3,>=2.21.0->tensorflow) (3.3.2) Requirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.10/site-packages (from requests<3,>=2.21.0->tensorflow) (3.7) Requirement already satisfied: urllib3<3,>=1.21.1 in /opt/conda/lib/python3.10/site-packages (from requests<3,>=2.21.0->tensorflow) (1.26.18) Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.10/site-packages (from requests<3,>=2.21.0->tensorflow) (2024.8.30) Requirement already satisfied: markdown>=2.6.8 in /opt/conda/lib/python3.10/site-packages (from tensorboard<2.17,>=2.16->tensorflow) (3.6) Requirement already satisfied: tensorboard-data-server<0.8.0,>=0.7.0 in /opt/conda/lib/python3.10/site-packages (from tensorboard<2.17,>=2.16->tensorflow) (0.7.2) Requirement already satisfied: werkzeug>=1.0.1 in /opt/conda/lib/python3.10/site-packages (from tensorboard<2.17,>=2.16->tensorflow) (3.0.4) Requirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /opt/conda/lib/python3.10/site-packages (from packaging->tensorflow) (3.1.2) Requirement already satisfied: MarkupSafe>=2.1.1 in /opt/conda/lib/python3.10/site-packages (from werkzeug>=1.0.1->tensorboard<2.17,>=2.16->tensorflow) (2.1.5) Requirement already satisfied: markdown-it-py>=2.2.0 in /opt/conda/lib/python3.10/site-packages (from rich->keras>=3.0.0->tensorflow) (3.0.0) Requirement already satisfied: pygments<3.0.0,>=2.13.0 in /opt/conda/lib/python3.10/site-packages (from rich->keras>=3.0.0->tensorflow) (2.18.0) Requirement already satisfied: mdurl~=0.1 in /opt/conda/lib/python3.10/site-packages (from markdown-it-py>=2.2.0->rich->keras>=3.0.0->tensorflow) (0.1.2)
Importing the required modules¶
In [3]:
# Modules used for data handling and visualisation
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import seaborn as sns
import random as r
sns.set_style("whitegrid")
# Modules used for suppressing warnings
import warnings
warnings.filterwarnings('ignore')
# Modules used for dataset split
import splitfolders
import os
# Modules used for model training and transfer learning
import tensorflow as tf
from tensorflow.keras.layers import Dense,Flatten
from tensorflow.keras.models import Sequential
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.applications.resnet50 import ResNet50
from tensorflow.keras import Model
In [4]:
# Centering all the output images in the notebook.
from IPython.core.display import HTML as Center
Center(""" <style>
.output_png {
display: table-cell;
text-align: center;
vertical-align: middle;
}
</style> """)
Out[4]:
Dataset Exploration¶
In [5]:
class Dataset:
def __init__(self, dataset_path : str):
self.PARENT = dataset_path
self.class_distribution = dict()
def __compute_class_distributions(self):
for dirname in os.listdir(self.PARENT):
self.class_distribution[dirname] = len(os.listdir(os.path.join(self.PARENT, dirname)))
def class_distributions(self):
self.__compute_class_distributions()
plt.figure(figsize=(10,10))
plt.bar(self.class_distribution.keys(),
self.class_distribution.values(),
color=["crimson","red","orange","yellow"])
plt.xticks(rotation=90)
plt.title("Class Distribution of PlantVillage dataset")
plt.xlabel("Class Label")
plt.ylabel("Frequency of class")
plt.show()
def show_class_samples(self):
rows = 5
columns = 3
c = 0
fig, axs = plt.subplots(rows, columns,figsize=(15,15))
for dirname in os.listdir(self.PARENT):
img_path = r.choice(os.listdir(os.path.join(self.PARENT, dirname)))
image = mpimg.imread(os.path.join(self.PARENT, dirname, img_path))
axs[c//columns, c%columns].imshow(image)
axs[c//columns, c%columns].set_title(dirname)
c += 1
fig.suptitle("Image Samples of Plant Village dataset")
plt.subplots_adjust(bottom=0.1, top=0.9, hspace=0.5)
plt.show()
Loading the dataset¶
In [8]:
plant_village = Dataset("/kaggle/input/plantdisease/PlantVillage")
Class Distribution¶
In [9]:
plant_village.class_distributions()
Sample Images¶
In [10]:
plant_village.show_class_samples()
Train, Test, Validation Split¶
In [11]:
class DataSplit:
def __init__(self, dataset_path : str, destination_path : str, train : float, test : float, val : float) -> None:
self.PARENT = dataset_path
self.TRAIN = train
self.TEST = test
self.VAL = val
self.destination_path = destination_path
self.train_gen = None
self.test_gen = None
self.val_gen = None
self.TRAIN_DIR = "dataset_new/train"
self.TEST_DIR = "dataset_new/test"
self.VAL_DIR = "dataset_new/val"
def test_train_validation_split(self):
#assert (self.TRAIN + self.TEST + self.VAL) == 1
splitfolders.ratio(input = self.PARENT,
output = self.destination_path,
seed = 1337, ratio = (.7, .1, .2),
group_prefix = None,
move = False)
def create_generators(self):
self.train_gen = tf.keras.preprocessing.image.ImageDataGenerator(
preprocessing_function=tf.keras.applications.resnet50.preprocess_input,
)
self.test_gen = tf.keras.preprocessing.image.ImageDataGenerator(
preprocessing_function=tf.keras.applications.resnet50.preprocess_input
)
self.val_gen = tf.keras.preprocessing.image.ImageDataGenerator(
preprocessing_function=tf.keras.applications.resnet50.preprocess_input
)
def get_images(self):
train_images = self.train_gen.flow_from_directory(
directory=self.TRAIN_DIR,
target_size=(75, 75),
color_mode='rgb',
class_mode='categorical',
batch_size=32,
shuffle=True,
seed=42,
subset='training'
)
val_images = self.val_gen.flow_from_directory(
directory=self.VAL_DIR,
target_size=(75, 75),
color_mode='rgb',
class_mode='categorical',
batch_size=32,
shuffle=True,
seed=42
)
test_images = self.test_gen.flow_from_directory(
directory=self.TEST_DIR,
target_size=(75, 75),
color_mode='rgb',
class_mode='categorical',
batch_size=32,
shuffle=False,
seed=42
)
return train_images, val_images, test_images
In [14]:
ds = DataSplit("/kaggle/input/plantdisease/PlantVillage","dataset_new",0.7,0.2, 0.1)
In [15]:
ds.test_train_validation_split()
Copying files: 20639 files [03:43, 92.21 files/s]
Train Data Insights¶
In [16]:
train = Dataset("dataset_new/train/")
In [17]:
train.class_distributions()
In [18]:
train.show_class_samples()
Test Data Insights¶
In [19]:
test = Dataset("dataset_new/test/")
In [20]:
test.class_distributions()
In [21]:
test.show_class_samples()
Validation Data Insights¶
In [22]:
val = Dataset("dataset_new/val/")
In [23]:
val.class_distributions()
In [24]:
val.show_class_samples()
Creating the data generators¶
In [25]:
ds.create_generators()
In [26]:
train, val, test = ds.get_images()
Found 14440 images belonging to 15 classes. Found 2058 images belonging to 15 classes. Found 4140 images belonging to 15 classes.
Transfer Learning¶
In [27]:
class TransferLearning:
def __init__(self, train, val) -> None:
self.train = train
self.val = val
self.model = None
self.history = None
def load_model(self):
self.model = ResNet50(weights = 'imagenet',
include_top = False,
input_shape = (75,75,3))
def mark_layers_non_trainable(self):
for layer in self.model.layers:
layer.trainable = False
def add_final_layer(self):
self.x = Flatten()(self.model.output)
self.x = Dense(1000, activation='relu')(self.x)
self.predictions = Dense(15, activation = 'softmax')(self.x)
def compile_model(self):
self.model = Model(inputs = self.model.input, outputs = self.predictions)
self.model.compile(optimizer='adam', loss="categorical_crossentropy", metrics=['accuracy'])
def train_model(self):
self.history = self.model.fit(train,
batch_size=32,
epochs=10, validation_data=val)
def plot_history(self):
fig, axs = plt.subplots(2, 1, figsize=(15,15))
axs[0].plot(self.history.history['loss'])
axs[0].plot(self.history.history['val_loss'])
axs[0].title.set_text('Training Loss vs Validation Loss')
axs[0].set_xlabel('Epochs')
axs[0].set_ylabel('Loss')
axs[0].legend(['Train','Val'])
axs[1].plot(self.history.history['accuracy'])
axs[1].plot(self.history.history['val_accuracy'])
axs[1].title.set_text('Training Accuracy vs Validation Accuracy')
axs[1].set_xlabel('Epochs')
axs[1].set_ylabel('Accuracy')
axs[1].legend(['Train', 'Val'])
Transfer Learning using Resnet50¶
In [28]:
tl = TransferLearning(train=train, val=val)
Loading the Resnet50 from Keras Application¶
In [29]:
tl.load_model()
Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/resnet/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5 94765736/94765736 ━━━━━━━━━━━━━━━━━━━━ 4s 0us/step
Making all the layers of the model non-trainable¶
In [30]:
tl.mark_layers_non_trainable()
Adding a final layer for classification of 15 classes¶
In [31]:
tl.add_final_layer()
Compiling model¶
In [32]:
tl.compile_model()
Training model¶
In [33]:
tl.train_model()
Epoch 1/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 292s 627ms/step - accuracy: 0.6837 - loss: 4.6579 - val_accuracy: 0.8780 - val_loss: 0.3906 Epoch 2/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 282s 623ms/step - accuracy: 0.9300 - loss: 0.1974 - val_accuracy: 0.8776 - val_loss: 0.4215 Epoch 3/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 323s 625ms/step - accuracy: 0.9592 - loss: 0.1229 - val_accuracy: 0.8946 - val_loss: 0.3885 Epoch 4/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 283s 627ms/step - accuracy: 0.9695 - loss: 0.0903 - val_accuracy: 0.8499 - val_loss: 0.7065 Epoch 5/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 282s 623ms/step - accuracy: 0.9691 - loss: 0.0973 - val_accuracy: 0.9062 - val_loss: 0.4080 Epoch 6/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 282s 624ms/step - accuracy: 0.9712 - loss: 0.0931 - val_accuracy: 0.9014 - val_loss: 0.4127 Epoch 7/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 281s 622ms/step - accuracy: 0.9812 - loss: 0.0627 - val_accuracy: 0.8950 - val_loss: 0.5721 Epoch 8/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 282s 622ms/step - accuracy: 0.9721 - loss: 0.0987 - val_accuracy: 0.8999 - val_loss: 0.5684 Epoch 9/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 282s 624ms/step - accuracy: 0.9842 - loss: 0.0485 - val_accuracy: 0.8746 - val_loss: 0.7685 Epoch 10/10 452/452 ━━━━━━━━━━━━━━━━━━━━ 285s 630ms/step - accuracy: 0.9790 - loss: 0.0743 - val_accuracy: 0.9130 - val_loss: 0.4779
In [34]:
tl.model.save("models/first_model_split_new.h5")
In [35]:
CLASS_NAMES = list(train.class_indices.keys())
CLASS_NAMES
Out[35]:
['Pepper__bell___Bacterial_spot', 'Pepper__bell___healthy', 'Potato___Early_blight', 'Potato___Late_blight', 'Potato___healthy', 'Tomato_Bacterial_spot', 'Tomato_Early_blight', 'Tomato_Late_blight', 'Tomato_Leaf_Mold', 'Tomato_Septoria_leaf_spot', 'Tomato_Spider_mites_Two_spotted_spider_mite', 'Tomato__Target_Spot', 'Tomato__Tomato_YellowLeaf__Curl_Virus', 'Tomato__Tomato_mosaic_virus', 'Tomato_healthy']
In [36]:
from sklearn.metrics import accuracy_score, classification_report
In [37]:
predictions = np.argmax(tl.model.predict(test), axis=1)
130/130 ━━━━━━━━━━━━━━━━━━━━ 55s 404ms/step
In [38]:
acc = accuracy_score(test.labels, predictions)
cm = tf.math.confusion_matrix(test.labels, predictions)
clr = classification_report(test.labels, predictions, target_names=CLASS_NAMES)
print("Test Accuracy: {:.3f}%".format(acc * 100))
plt.figure(figsize=(8, 8))
sns.heatmap(cm, annot=True, fmt='g', vmin=0, cmap='Blues', cbar=False)
plt.xticks(ticks= np.arange(15) + 0.5, labels=CLASS_NAMES, rotation=90)
plt.yticks(ticks= np.arange(15) + 0.5, labels=CLASS_NAMES, rotation=0)
plt.xlabel("Predicted")
plt.ylabel("Actual")
plt.title("Confusion Matrix")
plt.show()
Test Accuracy: 90.700%
In [39]:
print(clr)
precision recall f1-score support
Pepper__bell___Bacterial_spot 0.91 0.96 0.93 201
Pepper__bell___healthy 0.96 0.98 0.97 297
Potato___Early_blight 0.90 0.97 0.94 200
Potato___Late_blight 0.87 0.86 0.87 200
Potato___healthy 0.96 0.74 0.84 31
Tomato_Bacterial_spot 0.95 0.92 0.94 427
Tomato_Early_blight 0.86 0.65 0.74 200
Tomato_Late_blight 0.82 0.89 0.86 383
Tomato_Leaf_Mold 0.99 0.78 0.87 191
Tomato_Septoria_leaf_spot 0.83 0.94 0.88 355
Tomato_Spider_mites_Two_spotted_spider_mite 0.95 0.83 0.89 336
Tomato__Target_Spot 0.82 0.89 0.85 282
Tomato__Tomato_YellowLeaf__Curl_Virus 0.97 0.97 0.97 643
Tomato__Tomato_mosaic_virus 0.90 0.95 0.92 75
Tomato_healthy 0.94 0.98 0.96 319
accuracy 0.91 4140
macro avg 0.91 0.89 0.89 4140
weighted avg 0.91 0.91 0.91 4140
Plotting the Learning Curves¶
In [40]:
tl.plot_history()
- We can observe that model achieves an accuracy of 97.50% and 90.43% on training and validation sets respectively.
- Moreover, we can also gauge that the model is overfitting slightly which can be handled by fine tuning the model using regularization and re-training the layers
Fine-tuning¶
Fine Tuning is the approach in which a pretrained model is used. However, few of the layers are made trainable to understand the patterns in the current dataset. Morevoer, regularization can also be added in the form of dropout layers.
In [41]:
class FineTuning:
def __init__(self, train, val) -> None:
self.train = train
self.val = val
self.model = None
self.history = None
self.fine_tune_from = 100
def load_model(self):
self.model = ResNet50(weights = 'imagenet',
include_top = False,
input_shape = (75,75,3))
def fine_tune(self):
for layer in self.model.layers[:self.fine_tune_from]:
layer.trainable = False
for layer in self.model.layers[self.fine_tune_from:]:
layer.trainable = True
def add_final_layer(self):
self.x = Flatten()(self.model.output)
self.x = Dense(1000, activation='relu')(self.x)
self.predictions = Dense(15, activation = 'softmax')(self.x)
def compile_model(self):
self.model = Model(inputs = self.model.input, outputs = self.predictions)
self.model.compile(optimizer='adam', loss="categorical_crossentropy", metrics=['accuracy'])
def train_model(self):
self.history = self.model.fit(train,
batch_size=32,
epochs=5,
validation_data=val,
callbacks=[
tf.keras.callbacks.EarlyStopping(
monitor='val_loss',
patience=3,
restore_best_weights=True
)
])
def plot_history(self):
fig, axs = plt.subplots(2, 1, figsize=(15,15))
axs[0].plot(self.history.history['loss'])
axs[0].plot(self.history.history['val_loss'])
axs[0].title.set_text('Training Loss vs Validation Loss')
axs[0].set_xlabel('Epochs')
axs[0].set_ylabel('Loss')
axs[0].legend(['Train','Val'])
axs[1].plot(self.history.history['accuracy'])
axs[1].plot(self.history.history['val_accuracy'])
axs[1].title.set_text('Training Accuracy vs Validation Accuracy')
axs[1].set_xlabel('Epochs')
axs[1].set_ylabel('Accuracy')
axs[1].legend(['Train', 'Val'])
Fine Tuning the ResNet50 model¶
In [42]:
ft = FineTuning(train,val)
Loading the ResNet50 model from keras applications¶
In [43]:
ft.load_model()
Making last 75 layers of the ResNet50 model trainable¶
In [44]:
ft.fine_tune()
Adding a final layer for classification of 15 classes¶
In [45]:
ft.add_final_layer()
Compiling the model¶
In [46]:
ft.compile_model()
Training the model for 5 epochs¶
In [47]:
ft.train_model()
Epoch 1/5 452/452 ━━━━━━━━━━━━━━━━━━━━ 738s 2s/step - accuracy: 0.7337 - loss: 1.7488 - val_accuracy: 0.9310 - val_loss: 0.2338 Epoch 2/5 452/452 ━━━━━━━━━━━━━━━━━━━━ 708s 2s/step - accuracy: 0.9514 - loss: 0.1604 - val_accuracy: 0.9475 - val_loss: 0.1709 Epoch 3/5 452/452 ━━━━━━━━━━━━━━━━━━━━ 710s 2s/step - accuracy: 0.9725 - loss: 0.0948 - val_accuracy: 0.8863 - val_loss: 1.5194 Epoch 4/5 452/452 ━━━━━━━━━━━━━━━━━━━━ 705s 2s/step - accuracy: 0.9785 - loss: 0.0780 - val_accuracy: 0.9538 - val_loss: 0.1580 Epoch 5/5 452/452 ━━━━━━━━━━━━━━━━━━━━ 708s 2s/step - accuracy: 0.9821 - loss: 0.0734 - val_accuracy: 0.9495 - val_loss: 0.1692
In [49]:
ft.model.save("models/second_model_split_new.h5")
In [50]:
predictions = np.argmax(ft.model.predict(test), axis=1)
130/130 ━━━━━━━━━━━━━━━━━━━━ 54s 397ms/step
In [51]:
acc = accuracy_score(test.labels, predictions)
cm = tf.math.confusion_matrix(test.labels, predictions)
clr = classification_report(test.labels, predictions, target_names=CLASS_NAMES)
print("Test Accuracy: {:.3f}%".format(acc * 100))
plt.figure(figsize=(8, 8))
sns.heatmap(cm, annot=True, fmt='g', vmin=0, cmap='Blues', cbar=False)
plt.xticks(ticks= np.arange(15) + 0.5, labels=CLASS_NAMES, rotation=90)
plt.yticks(ticks= np.arange(15) + 0.5, labels=CLASS_NAMES, rotation=0)
plt.xlabel("Predicted")
plt.ylabel("Actual")
plt.title("Confusion Matrix")
plt.show()
Test Accuracy: 95.338%
In [52]:
print(clr)
precision recall f1-score support
Pepper__bell___Bacterial_spot 0.92 0.98 0.94 201
Pepper__bell___healthy 0.99 0.97 0.98 297
Potato___Early_blight 0.93 0.98 0.96 200
Potato___Late_blight 0.94 0.94 0.94 200
Potato___healthy 0.88 0.97 0.92 31
Tomato_Bacterial_spot 0.95 0.99 0.97 427
Tomato_Early_blight 0.98 0.81 0.88 200
Tomato_Late_blight 0.94 0.95 0.95 383
Tomato_Leaf_Mold 0.96 0.96 0.96 191
Tomato_Septoria_leaf_spot 0.95 0.96 0.95 355
Tomato_Spider_mites_Two_spotted_spider_mite 0.89 0.99 0.94 336
Tomato__Target_Spot 0.95 0.86 0.91 282
Tomato__Tomato_YellowLeaf__Curl_Virus 1.00 0.95 0.97 643
Tomato__Tomato_mosaic_virus 0.96 0.99 0.97 75
Tomato_healthy 0.97 0.99 0.98 319
accuracy 0.95 4140
macro avg 0.95 0.95 0.95 4140
weighted avg 0.95 0.95 0.95 4140
Evaluation of the fine-tuned model¶
In [53]:
ft.model.evaluate(test)
130/130 ━━━━━━━━━━━━━━━━━━━━ 50s 383ms/step - accuracy: 0.9586 - loss: 0.1396
Out[53]:
[0.16538484394550323, 0.9533816576004028]
Plotting the learning curves of the fine-tuning process¶
In [54]:
ft.plot_history()
Conclusion¶
- Transfer Learning is approach of using a model pretrained(i.e. ResNet50) on a large dataset(here, imagenet) and using its knowledge for our case.
- As inferred earlier, transfer learning gives an accuracy of 97.50% and 90.43% on training and validation sets respectively which shows that the model is slightly overfitted resulting in the requirement of fine tuning of the model.
- The model is fine tuned by letting the last 75 layers learn the patterns in the dataset and overcome the overfitting and improve the accuracy.
- The fine tuned model gives an accuracy of 98.09%, 94.31%, and 95.23% on train, validation, and test splits.
- On a final note, in the deep learning there is required of the large overhead of time and hardware requirements for Fine Tuning of the model.